Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Department of Human Genetics, University of Chicago, Chicago, IL 60637; bSainte-Justine Hospital Research Centre, Department of Pediatrics, Faculty of
Medicine, University of Montreal, Montreal H3T 1C5, QC, Canada; cUnit of Mycobacterial Genetics, Institut Pasteur, Paris 75015, France; and dEuropean
Bioinformatics Institute, Wellcome Trust Genome Campus, Hinxton, Cambridge CB10 1SD, United Kingdom
Edited by Barry R. Bloom, Harvard School of Public Health, Boston, MA, and approved December 15, 2011 (received for review September 24, 2011)
from 65 healthy individuals with a virulent strain of MTB. Following infection, we extracted RNA from the untreated and
infected DCs at the same time, 18 h after the infection. We then
characterized genome-wide gene expression proles in all samples using the Illumina HT-12 expression arrays. After excluding
data from poorly annotated array probes and from genes that
were classied as not expressed, we normalized the expression
data for the remaining 12,958 genes (details in Materials and
Author contributions: L.B.B., L.T., and Y.G. designed research; L.B.B. and L.T. performed
research; L.T., A.A.P., B.G., J.C.M., and Y.G. contributed new reagents/analytic tools; L.B.B.
analyzed data; and L.B.B. and Y.G. wrote the paper.
The authors declare no conict of interest.
This article is a PNAS Direct Submission.
Freely available online through the PNAS open access option.
Data deposition: The gene expression and genotype data reported in this paper have
been deposited in the Gene Expression Omnibus (GEO) database (accession nos. GSE34151
and GSE34588, respectively).
1
www.pnas.org/cgi/doi/10.1073/pnas.1115761109
GENETICS
genes whose expression levels were altered following MTB infection were 1.6 times more likely to be associated with cis-eQTL
than the genome-wide average (14% compared with the 9% that
are expected by chance alone; P = 1.6 1016; Fig. 2B). Our
results thus suggest that genes involved in immune response to
MTB have increased levels of functional diversity in the population. Such functional variation is likely to underlie interindividual
differences in susceptibility to infectious diseases in general and
MTB infection in particular.
As expected, there is a large overlap of eQTL identied independently in the noninfected and infected DCs (Pearson
correlation of eQTL association P values in the two classes of
DCs is 0.68; P < 1015). In the context of susceptibility to TB,
however, the most interesting eQTL are arguably those with
a different effect on gene expression levels before and after infection with MTB, which we term response eQTL. Response
eQTL likely interact with MTB (directly or indirectly) and thus
may account for interindividual variation in immune response to
MTB infection. We classied response eQTL by using highly
conservative criteria to minimize the probability of false positives
(e.g., when true eQTL in both untreated and infected DCs are
only classied as such in one class because of incomplete power).
Specically, we dened response eQTL when we found strong
evidence for a cis-eQTL for a gene in either untreated or
infected DCs at an FDR of 1% (P < 1.4 105) and no statistical
evidence supporting a cis-eQTL for the same gene (in the entire
tested 200-kb region) in the other condition at a very relaxed
FDR threshold of 50% (more details in Materials and Methods).
Using this approach, we identied 198 genes with strong evidence of being associated with at least one response eQTL (102
and 96 eQTL in untreated and infected DCs, respectively; Fig. 2
A, C, and D and Dataset S3).
Fig. 1. Functional characterization of immune responses to MTB infection. (A) Volcano plot showing differentially expressed genes after infection of DCs
with MTB for 18 h. The negative log10 transformed P values test the null hypothesis of no difference in expression levels between untreated and infected DCs
(y axis) and are plotted against the average log2 fold changes in expression (x axis). Data for genes that were not classied as differentially expressed are
plotted in black. In gray and blue, we plotted data for genes that are differentially expressed after infection with MTB (P value <7.7 107; Bonferroni
corrected P value <0.01) with an absolute log2 fold change (|FC|) less than or equal to 0.5 or greater than 0.5, respectively. (B) Gene ontology (GO) enrichment
analysis for genes that were classied as up- (red) or down- (blue) regulated following infection of DCs with MTB. Only signicant enrichments at an FDR <1%
are plotted (complete results in Dataset S2).
Barreiro et al.
Fig. 2. Deciphering the genetic basis of interindividual variation in immune response to MTB infection. (A) Plot contrasting the evidence for cis-eQTL in the untreated and
infected DCs. For every gene we plotted the additive model
P values (log10 transformed) for the most strongly associated
cis-SNP (dened as SNPs located in 200-kb window centered on
the TSS of a proximal gene) with gene expression levels in the
untreated (x axis) or infected (y axis) DCs, respectively. The red
dashed lines specify the P values corresponding to an FDR of
1%. The blue dashed lines specify the second, more relaxed,
cutoff (50% FDR) used to condently classify response eQTL.
Only genes with strong evidence of a cis-eQTL in at least one of
the conditions (FDR of 1%) are plotted. (B) Proportion of ciseQTL (y axis) observed among all tested genes and among
genes that were classied as differentially expressed (DEG)
following infection with MTB. (C) Example of a response eQTL
found only in the untreated samples. (D) Example of a response eQTL found only in the infected samples.
role in protective immunity against TB (2426) (Fig. 3B), including TNF- (460-fold induction), IFN- (24-fold induction),
and IL-12 (8-fold induction).
Using the protein secretion data we also looked for protein
QTL (pQTL), namely loci that are associated with cytokines/
chemokines secretion levels measured after infection of the
DCs. We did not nd any signicant association, either genomewide or when we restricted the analysis to SNPs within 200 kb of
the TSSs of the genes encoding the measured proteins. However,
when we only considered the SNPs that were previously identied as cis-eQTL (in the analysis of transcript levels above) we
found a clear enrichment in the association of genotypes with
1206 | www.pnas.org/cgi/doi/10.1073/pnas.1115761109
Barreiro et al.
Discussion
We studied variation in the regulatory response to MTB infection of DCs. We chose to focus on DCs because they have
been shown to play an essential, nonredundant role in protective
GENETICS
tantly, we did not observe a similar enrichment when we considered eQTL that are common to untreated and infected DCs (Fig.
4B; this observation serves as a control for a possible bias in the
power to detect an association in the GWAS when only eQTL
are considered).
We note that our observations are robust to the particular
method used to identify response eQTL. Indeed, an alternative
approach used to identify cis-interactions (i.e., response-eQTL)
is to consider variation in the magnitude of the shifts in gene
expression after a treatment, in our case MTB infection, as the
quantitative trait to be mapped (33, 34). Supporting our previous
ndings, MTB-response eQTL identied using this approach
(Dataset S3) were also signicantly enriched for GWAS P values
<0.05 (1.8-fold enrichment, 2 test, P = 0.01; Fig. S4).
Taken together, our results indicate that response eQTL are
likely enriched for susceptibility alleles for TB. In particular, our
data strongly support the notion that dual-specicity phosphatase 14 (DUSP14) is a new susceptibility gene for TB. Indeed, we
found strong evidence that rs712039 is an immune response
eQTL associated with variation in the expression levels of
DUSP14 exclusively in noninfected DCs (Pnoninfected DCs = 9.6
106 (FDR = 6 103); PMTB-infected DCs = 0.09 (FDR = 0.9);
Fig. 5A). In addition, in our focused GWAS analysis, which was
restricted to the set of SNPs we classied as immune response
eQTL (as the set of most likely candidates), the strength of the
association between rs712039 and pulmonary TB is signicant
even after a conservative Bonferroni correction (Fig. 4D). This
SNP was not discussed in the original GWAS paper, probably
because it was not signicant at a genome-wide threshold, yet it
shows one of the strongest genetic associations with pulmonary
TB, in both the Ghana cohort (P = 9.8 104) and the Gambia
replication cohort (P = 2.3 103, combined P = 3.3 106;
Table S1).
Fig. 4. MTB-response eQTL are strong candidates to impact susceptibility to pulmonary TB. (A) The median GWAS P value for an expanding window of genes
is plotted. We used the GWAS P values obtained when combining the Ghana and Gambia cohorts (16). Genes are ordered by the strength of evidence
supporting an association with an eQTL only in the untreated (Left) or the infected (Right) DCs, respectively. To avoid positional biases, we restricted our
analyses to the set of cis-SNPs that was tested in our study (i.e., SNPs located in 200-kb window centered on the TSS of proximal genes). (B) Histogram of the
proportion of GWAS SNPs with nominal P values <0.05 among all GWAS SNPs (gray), among SNPs that were classied as eQTL in both untreated and infected
DCs (blue), and among response eQTL (red). (C) Manhattan plot showing the negative log10 transformed P values (y axis) for the association between the
response eQTL identied in this study and susceptibility to pulmonary TB. The dashed line corresponds to the genome-wide signicance cutoff after
a conservative Bonferroni correction.
Barreiro et al.
Fig. 5. Genes with response eQTL likely to impact susceptibility to pulmonary TB. (A) Response eQTL identied for DUSP14. (B) Response eQTL
identied for RIPK2. (C) Response eQTL identied for ATP6V0A2.
1208 | www.pnas.org/cgi/doi/10.1073/pnas.1115761109
Barreiro et al.
GENETICS